linkedin-content

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.80). The SKILL.md Quick Start explicitly shows running "belt app run tavily/search-assistant" with a query for "LinkedIn viral post examples…" — a step that fetches open/web and user-generated content the agent is expected to read and use to craft posts, so third-party content can materially influence its actions.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly requires and uses the inference.sh CLI (https://inference.sh) — e.g., "belt app run ..." — which at runtime invokes remote apps (tavily/search-assistant, falai/flux-dev-lora, etc.) that fetch and execute remote app logic/prompts, so external content from inference.sh directly controls agent behavior.